1 /*
2 * Copyright (c) 2003-2004 Fabrice Bellard
3 * Copyright (c) 2019, 2024 Red Hat, Inc.
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a copy
6 * of this software and associated documentation files (the "Software"), to deal
7 * in the Software without restriction, including without limitation the rights
8 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
9 * copies of the Software, and to permit persons to whom the Software is
10 * furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice shall be included in
13 * all copies or substantial portions of the Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
21 * THE SOFTWARE.
22 */
23 #include "qemu/osdep.h"
24 #include "qemu/error-report.h"
25 #include "qemu/cutils.h"
26 #include "qemu/units.h"
27 #include "qemu/datadir.h"
28 #include "qapi/error.h"
29 #include "system/numa.h"
30 #include "system/system.h"
31 #include "system/xen.h"
32 #include "trace.h"
33
34 #include "hw/i386/x86.h"
35 #include "target/i386/cpu.h"
36 #include "hw/rtc/mc146818rtc.h"
37 #include "target/i386/sev.h"
38
39 #include "hw/acpi/cpu_hotplug.h"
40 #include "hw/irq.h"
41 #include "hw/loader.h"
42 #include "multiboot.h"
43 #include "elf.h"
44 #include "standard-headers/asm-x86/bootparam.h"
45 #include CONFIG_DEVICES
46 #include "kvm/kvm_i386.h"
47 #include "kvm/tdx.h"
48
49 #ifdef CONFIG_XEN_EMU
50 #include "hw/xen/xen.h"
51 #include "hw/i386/kvm/xen_evtchn.h"
52 #endif
53
54 /* Physical Address of PVH entry point read from kernel ELF NOTE */
55 static size_t pvh_start_addr;
56
x86_cpu_new(X86MachineState * x86ms,int64_t apic_id,Error ** errp)57 static void x86_cpu_new(X86MachineState *x86ms, int64_t apic_id, Error **errp)
58 {
59 Object *cpu = object_new(MACHINE(x86ms)->cpu_type);
60
61 if (!object_property_set_uint(cpu, "apic-id", apic_id, errp)) {
62 goto out;
63 }
64 qdev_realize(DEVICE(cpu), NULL, errp);
65
66 out:
67 object_unref(cpu);
68 }
69
x86_cpus_init(X86MachineState * x86ms,int default_cpu_version)70 void x86_cpus_init(X86MachineState *x86ms, int default_cpu_version)
71 {
72 int i;
73 const CPUArchIdList *possible_cpus;
74 MachineState *ms = MACHINE(x86ms);
75 MachineClass *mc = MACHINE_GET_CLASS(x86ms);
76
77 x86_cpu_set_default_version(default_cpu_version);
78
79 /*
80 * Calculates the limit to CPU APIC ID values
81 *
82 * Limit for the APIC ID value, so that all
83 * CPU APIC IDs are < x86ms->apic_id_limit.
84 *
85 * This is used for FW_CFG_MAX_CPUS. See comments on fw_cfg_arch_create().
86 */
87 x86ms->apic_id_limit = x86_cpu_apic_id_from_index(x86ms,
88 ms->smp.max_cpus - 1) + 1;
89
90 /*
91 * Can we support APIC ID 255 or higher? With KVM, that requires
92 * both in-kernel lapic and X2APIC userspace API.
93 *
94 * kvm_enabled() must go first to ensure that kvm_* references are
95 * not emitted for the linker to consume (kvm_enabled() is
96 * a literal `0` in configurations where kvm_* aren't defined)
97 */
98 if (kvm_enabled() && x86ms->apic_id_limit > 255 &&
99 kvm_irqchip_in_kernel() && !kvm_enable_x2apic()) {
100 error_report("current -smp configuration requires kernel "
101 "irqchip and X2APIC API support.");
102 exit(EXIT_FAILURE);
103 }
104
105 if (kvm_enabled()) {
106 kvm_set_max_apic_id(x86ms->apic_id_limit);
107 }
108
109 if (!kvm_irqchip_in_kernel()) {
110 apic_set_max_apic_id(x86ms->apic_id_limit);
111 }
112
113 possible_cpus = mc->possible_cpu_arch_ids(ms);
114 for (i = 0; i < ms->smp.cpus; i++) {
115 x86_cpu_new(x86ms, possible_cpus->cpus[i].arch_id, &error_fatal);
116 }
117 }
118
x86_rtc_set_cpus_count(ISADevice * s,uint16_t cpus_count)119 void x86_rtc_set_cpus_count(ISADevice *s, uint16_t cpus_count)
120 {
121 MC146818RtcState *rtc = MC146818_RTC(s);
122
123 if (cpus_count > 0xff) {
124 /*
125 * If the number of CPUs can't be represented in 8 bits, the
126 * BIOS must use "FW_CFG_NB_CPUS". Set RTC field to 0 just
127 * to make old BIOSes fail more predictably.
128 */
129 mc146818rtc_set_cmos_data(rtc, 0x5f, 0);
130 } else {
131 mc146818rtc_set_cmos_data(rtc, 0x5f, cpus_count - 1);
132 }
133 }
134
x86_apic_cmp(const void * a,const void * b)135 static int x86_apic_cmp(const void *a, const void *b)
136 {
137 CPUArchId *apic_a = (CPUArchId *)a;
138 CPUArchId *apic_b = (CPUArchId *)b;
139
140 return apic_a->arch_id - apic_b->arch_id;
141 }
142
143 /*
144 * returns pointer to CPUArchId descriptor that matches CPU's apic_id
145 * in ms->possible_cpus->cpus, if ms->possible_cpus->cpus has no
146 * entry corresponding to CPU's apic_id returns NULL.
147 */
x86_find_cpu_slot(MachineState * ms,uint32_t id,int * idx)148 static CPUArchId *x86_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
149 {
150 CPUArchId apic_id, *found_cpu;
151
152 apic_id.arch_id = id;
153 found_cpu = bsearch(&apic_id, ms->possible_cpus->cpus,
154 ms->possible_cpus->len, sizeof(*ms->possible_cpus->cpus),
155 x86_apic_cmp);
156 if (found_cpu && idx) {
157 *idx = found_cpu - ms->possible_cpus->cpus;
158 }
159 return found_cpu;
160 }
161
x86_cpu_plug(HotplugHandler * hotplug_dev,DeviceState * dev,Error ** errp)162 void x86_cpu_plug(HotplugHandler *hotplug_dev,
163 DeviceState *dev, Error **errp)
164 {
165 CPUArchId *found_cpu;
166 Error *local_err = NULL;
167 X86CPU *cpu = X86_CPU(dev);
168 X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
169
170 if (x86ms->acpi_dev) {
171 hotplug_handler_plug(x86ms->acpi_dev, dev, &local_err);
172 if (local_err) {
173 goto out;
174 }
175 }
176
177 /* increment the number of CPUs */
178 x86ms->boot_cpus++;
179 if (x86ms->rtc) {
180 x86_rtc_set_cpus_count(x86ms->rtc, x86ms->boot_cpus);
181 }
182 if (x86ms->fw_cfg) {
183 fw_cfg_modify_i16(x86ms->fw_cfg, FW_CFG_NB_CPUS, x86ms->boot_cpus);
184 }
185
186 /*
187 * Non-hotplugged CPUs get their SMM cpu address space initialized in
188 * machine init done notifier: register_smram_listener().
189 *
190 * We need initialize the SMM cpu address space for the hotplugged CPU
191 * specifically.
192 */
193 if (kvm_enabled() && dev->hotplugged && x86_machine_is_smm_enabled(x86ms)) {
194 kvm_smm_cpu_address_space_init(cpu);
195 }
196
197 found_cpu = x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, NULL);
198 found_cpu->cpu = CPU(dev);
199 out:
200 error_propagate(errp, local_err);
201 }
202
x86_cpu_unplug_request_cb(HotplugHandler * hotplug_dev,DeviceState * dev,Error ** errp)203 void x86_cpu_unplug_request_cb(HotplugHandler *hotplug_dev,
204 DeviceState *dev, Error **errp)
205 {
206 int idx = -1;
207 X86CPU *cpu = X86_CPU(dev);
208 X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
209
210 if (!x86ms->acpi_dev) {
211 error_setg(errp, "CPU hot unplug not supported without ACPI");
212 return;
213 }
214
215 x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, &idx);
216 assert(idx != -1);
217 if (idx == 0) {
218 error_setg(errp, "Boot CPU is unpluggable");
219 return;
220 }
221
222 hotplug_handler_unplug_request(x86ms->acpi_dev, dev,
223 errp);
224 }
225
x86_cpu_unplug_cb(HotplugHandler * hotplug_dev,DeviceState * dev,Error ** errp)226 void x86_cpu_unplug_cb(HotplugHandler *hotplug_dev,
227 DeviceState *dev, Error **errp)
228 {
229 CPUArchId *found_cpu;
230 Error *local_err = NULL;
231 X86CPU *cpu = X86_CPU(dev);
232 X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
233
234 hotplug_handler_unplug(x86ms->acpi_dev, dev, &local_err);
235 if (local_err) {
236 goto out;
237 }
238
239 found_cpu = x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, NULL);
240 found_cpu->cpu = NULL;
241 qdev_unrealize(dev);
242
243 /* decrement the number of CPUs */
244 x86ms->boot_cpus--;
245 /* Update the number of CPUs in CMOS */
246 x86_rtc_set_cpus_count(x86ms->rtc, x86ms->boot_cpus);
247 fw_cfg_modify_i16(x86ms->fw_cfg, FW_CFG_NB_CPUS, x86ms->boot_cpus);
248 out:
249 error_propagate(errp, local_err);
250 }
251
x86_cpu_pre_plug(HotplugHandler * hotplug_dev,DeviceState * dev,Error ** errp)252 void x86_cpu_pre_plug(HotplugHandler *hotplug_dev,
253 DeviceState *dev, Error **errp)
254 {
255 int idx;
256 CPUState *cs;
257 CPUArchId *cpu_slot;
258 X86CPUTopoIDs topo_ids;
259 X86CPU *cpu = X86_CPU(dev);
260 CPUX86State *env = &cpu->env;
261 MachineState *ms = MACHINE(hotplug_dev);
262 X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
263 X86CPUTopoInfo *topo_info = &env->topo_info;
264
265 if (!object_dynamic_cast(OBJECT(cpu), ms->cpu_type)) {
266 error_setg(errp, "Invalid CPU type, expected cpu type: '%s'",
267 ms->cpu_type);
268 return;
269 }
270
271 if (x86ms->acpi_dev) {
272 Error *local_err = NULL;
273
274 hotplug_handler_pre_plug(HOTPLUG_HANDLER(x86ms->acpi_dev), dev,
275 &local_err);
276 if (local_err) {
277 error_propagate(errp, local_err);
278 return;
279 }
280 }
281
282 init_topo_info(topo_info, x86ms);
283
284 if (ms->smp.modules > 1) {
285 set_bit(CPU_TOPOLOGY_LEVEL_MODULE, env->avail_cpu_topo);
286 }
287
288 if (ms->smp.dies > 1) {
289 set_bit(CPU_TOPOLOGY_LEVEL_DIE, env->avail_cpu_topo);
290 }
291
292 /*
293 * If APIC ID is not set,
294 * set it based on socket/die/module/core/thread properties.
295 */
296 if (cpu->apic_id == UNASSIGNED_APIC_ID) {
297 /*
298 * die-id was optional in QEMU 4.0 and older, so keep it optional
299 * if there's only one die per socket.
300 */
301 if (cpu->die_id < 0 && ms->smp.dies == 1) {
302 cpu->die_id = 0;
303 }
304
305 /*
306 * module-id was optional in QEMU 9.0 and older, so keep it optional
307 * if there's only one module per die.
308 */
309 if (cpu->module_id < 0 && ms->smp.modules == 1) {
310 cpu->module_id = 0;
311 }
312
313 if (cpu->socket_id < 0) {
314 error_setg(errp, "CPU socket-id is not set");
315 return;
316 } else if (cpu->socket_id > ms->smp.sockets - 1) {
317 error_setg(errp, "Invalid CPU socket-id: %u must be in range 0:%u",
318 cpu->socket_id, ms->smp.sockets - 1);
319 return;
320 }
321 if (cpu->die_id < 0) {
322 error_setg(errp, "CPU die-id is not set");
323 return;
324 } else if (cpu->die_id > ms->smp.dies - 1) {
325 error_setg(errp, "Invalid CPU die-id: %u must be in range 0:%u",
326 cpu->die_id, ms->smp.dies - 1);
327 return;
328 }
329 if (cpu->module_id < 0) {
330 error_setg(errp, "CPU module-id is not set");
331 return;
332 } else if (cpu->module_id > ms->smp.modules - 1) {
333 error_setg(errp, "Invalid CPU module-id: %u must be in range 0:%u",
334 cpu->module_id, ms->smp.modules - 1);
335 return;
336 }
337 if (cpu->core_id < 0) {
338 error_setg(errp, "CPU core-id is not set");
339 return;
340 } else if (cpu->core_id > (ms->smp.cores - 1)) {
341 error_setg(errp, "Invalid CPU core-id: %u must be in range 0:%u",
342 cpu->core_id, ms->smp.cores - 1);
343 return;
344 }
345 if (cpu->thread_id < 0) {
346 error_setg(errp, "CPU thread-id is not set");
347 return;
348 } else if (cpu->thread_id > (ms->smp.threads - 1)) {
349 error_setg(errp, "Invalid CPU thread-id: %u must be in range 0:%u",
350 cpu->thread_id, ms->smp.threads - 1);
351 return;
352 }
353
354 topo_ids.pkg_id = cpu->socket_id;
355 topo_ids.die_id = cpu->die_id;
356 topo_ids.module_id = cpu->module_id;
357 topo_ids.core_id = cpu->core_id;
358 topo_ids.smt_id = cpu->thread_id;
359 cpu->apic_id = x86_apicid_from_topo_ids(topo_info, &topo_ids);
360 }
361
362 cpu_slot = x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, &idx);
363 if (!cpu_slot) {
364 x86_topo_ids_from_apicid(cpu->apic_id, topo_info, &topo_ids);
365
366 error_setg(errp,
367 "Invalid CPU [socket: %u, die: %u, module: %u, core: %u, thread: %u]"
368 " with APIC ID %" PRIu32 ", valid index range 0:%d",
369 topo_ids.pkg_id, topo_ids.die_id, topo_ids.module_id,
370 topo_ids.core_id, topo_ids.smt_id, cpu->apic_id,
371 ms->possible_cpus->len - 1);
372 return;
373 }
374
375 if (cpu_slot->cpu) {
376 error_setg(errp, "CPU[%d] with APIC ID %" PRIu32 " exists",
377 idx, cpu->apic_id);
378 return;
379 }
380
381 /* if 'address' properties socket-id/core-id/thread-id are not set, set them
382 * so that machine_query_hotpluggable_cpus would show correct values
383 */
384 /* TODO: move socket_id/core_id/thread_id checks into x86_cpu_realizefn()
385 * once -smp refactoring is complete and there will be CPU private
386 * CPUState::nr_cores and CPUState::nr_threads fields instead of globals */
387 x86_topo_ids_from_apicid(cpu->apic_id, topo_info, &topo_ids);
388 if (cpu->socket_id != -1 && cpu->socket_id != topo_ids.pkg_id) {
389 error_setg(errp, "property socket-id: %u doesn't match set apic-id:"
390 " 0x%x (socket-id: %u)", cpu->socket_id, cpu->apic_id,
391 topo_ids.pkg_id);
392 return;
393 }
394 cpu->socket_id = topo_ids.pkg_id;
395
396 if (cpu->die_id != -1 && cpu->die_id != topo_ids.die_id) {
397 error_setg(errp, "property die-id: %u doesn't match set apic-id:"
398 " 0x%x (die-id: %u)", cpu->die_id, cpu->apic_id, topo_ids.die_id);
399 return;
400 }
401 cpu->die_id = topo_ids.die_id;
402
403 if (cpu->module_id != -1 && cpu->module_id != topo_ids.module_id) {
404 error_setg(errp, "property module-id: %u doesn't match set apic-id:"
405 " 0x%x (module-id: %u)", cpu->module_id, cpu->apic_id,
406 topo_ids.module_id);
407 return;
408 }
409 cpu->module_id = topo_ids.module_id;
410
411 if (cpu->core_id != -1 && cpu->core_id != topo_ids.core_id) {
412 error_setg(errp, "property core-id: %u doesn't match set apic-id:"
413 " 0x%x (core-id: %u)", cpu->core_id, cpu->apic_id,
414 topo_ids.core_id);
415 return;
416 }
417 cpu->core_id = topo_ids.core_id;
418
419 if (cpu->thread_id != -1 && cpu->thread_id != topo_ids.smt_id) {
420 error_setg(errp, "property thread-id: %u doesn't match set apic-id:"
421 " 0x%x (thread-id: %u)", cpu->thread_id, cpu->apic_id,
422 topo_ids.smt_id);
423 return;
424 }
425 cpu->thread_id = topo_ids.smt_id;
426
427 /*
428 * kvm_enabled() must go first to ensure that kvm_* references are
429 * not emitted for the linker to consume (kvm_enabled() is
430 * a literal `0` in configurations where kvm_* aren't defined)
431 */
432 if (kvm_enabled() && hyperv_feat_enabled(cpu, HYPERV_FEAT_VPINDEX) &&
433 !kvm_hv_vpindex_settable()) {
434 error_setg(errp, "kernel doesn't allow setting HyperV VP_INDEX");
435 return;
436 }
437
438 cs = CPU(cpu);
439 cs->cpu_index = idx;
440
441 numa_cpu_pre_plug(cpu_slot, dev, errp);
442 }
443
get_file_size(FILE * f)444 static long get_file_size(FILE *f)
445 {
446 long where, size;
447
448 /* XXX: on Unix systems, using fstat() probably makes more sense */
449
450 where = ftell(f);
451 fseek(f, 0, SEEK_END);
452 size = ftell(f);
453 fseek(f, where, SEEK_SET);
454
455 return size;
456 }
457
gsi_handler(void * opaque,int n,int level)458 void gsi_handler(void *opaque, int n, int level)
459 {
460 GSIState *s = opaque;
461 bool bypass_ioapic = false;
462
463 trace_x86_gsi_interrupt(n, level);
464
465 #ifdef CONFIG_XEN_EMU
466 /*
467 * Xen delivers the GSI to the Legacy PIC (not that Legacy PIC
468 * routing actually works properly under Xen). And then to
469 * *either* the PIRQ handling or the I/OAPIC depending on whether
470 * the former wants it.
471 *
472 * Additionally, this hook allows the Xen event channel GSI to
473 * work around QEMU's lack of support for shared level interrupts,
474 * by keeping track of the externally driven state of the pin and
475 * implementing a logical OR with the state of the evtchn GSI.
476 */
477 if (xen_mode == XEN_EMULATE) {
478 bypass_ioapic = xen_evtchn_set_gsi(n, &level);
479 }
480 #endif
481
482 switch (n) {
483 case 0 ... ISA_NUM_IRQS - 1:
484 if (s->i8259_irq[n]) {
485 /* Under KVM, Kernel will forward to both PIC and IOAPIC */
486 qemu_set_irq(s->i8259_irq[n], level);
487 }
488 /* fall through */
489 case ISA_NUM_IRQS ... IOAPIC_NUM_PINS - 1:
490 if (!bypass_ioapic) {
491 qemu_set_irq(s->ioapic_irq[n], level);
492 }
493 break;
494 case IO_APIC_SECONDARY_IRQBASE
495 ... IO_APIC_SECONDARY_IRQBASE + IOAPIC_NUM_PINS - 1:
496 qemu_set_irq(s->ioapic2_irq[n - IO_APIC_SECONDARY_IRQBASE], level);
497 break;
498 }
499 }
500
ioapic_init_gsi(GSIState * gsi_state,Object * parent)501 void ioapic_init_gsi(GSIState *gsi_state, Object *parent)
502 {
503 DeviceState *dev;
504 SysBusDevice *d;
505 unsigned int i;
506
507 assert(parent);
508 if (kvm_ioapic_in_kernel()) {
509 dev = qdev_new(TYPE_KVM_IOAPIC);
510 } else {
511 dev = qdev_new(TYPE_IOAPIC);
512 }
513 object_property_add_child(parent, "ioapic", OBJECT(dev));
514 d = SYS_BUS_DEVICE(dev);
515 sysbus_realize_and_unref(d, &error_fatal);
516 sysbus_mmio_map(d, 0, IO_APIC_DEFAULT_ADDRESS);
517
518 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
519 gsi_state->ioapic_irq[i] = qdev_get_gpio_in(dev, i);
520 }
521 }
522
ioapic_init_secondary(GSIState * gsi_state)523 DeviceState *ioapic_init_secondary(GSIState *gsi_state)
524 {
525 DeviceState *dev;
526 SysBusDevice *d;
527 unsigned int i;
528
529 dev = qdev_new(TYPE_IOAPIC);
530 d = SYS_BUS_DEVICE(dev);
531 sysbus_realize_and_unref(d, &error_fatal);
532 sysbus_mmio_map(d, 0, IO_APIC_SECONDARY_ADDRESS);
533
534 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
535 gsi_state->ioapic2_irq[i] = qdev_get_gpio_in(dev, i);
536 }
537 return dev;
538 }
539
540 /*
541 * The entry point into the kernel for PVH boot is different from
542 * the native entry point. The PVH entry is defined by the x86/HVM
543 * direct boot ABI and is available in an ELFNOTE in the kernel binary.
544 *
545 * This function is passed to load_elf() when it is called from
546 * load_elfboot() which then additionally checks for an ELF Note of
547 * type XEN_ELFNOTE_PHYS32_ENTRY and passes it to this function to
548 * parse the PVH entry address from the ELF Note.
549 *
550 * Due to trickery in elf_opts.h, load_elf() is actually available as
551 * load_elf32() or load_elf64() and this routine needs to be able
552 * to deal with being called as 32 or 64 bit.
553 *
554 * The address of the PVH entry point is saved to the 'pvh_start_addr'
555 * global variable. (although the entry point is 32-bit, the kernel
556 * binary can be either 32-bit or 64-bit).
557 */
read_pvh_start_addr(void * arg1,void * arg2,bool is64)558 static uint64_t read_pvh_start_addr(void *arg1, void *arg2, bool is64)
559 {
560 size_t *elf_note_data_addr;
561
562 /* Check if ELF Note header passed in is valid */
563 if (arg1 == NULL) {
564 return 0;
565 }
566
567 if (is64) {
568 struct elf64_note *nhdr64 = (struct elf64_note *)arg1;
569 uint64_t nhdr_size64 = sizeof(struct elf64_note);
570 uint64_t phdr_align = *(uint64_t *)arg2;
571 uint64_t nhdr_namesz = nhdr64->n_namesz;
572
573 elf_note_data_addr =
574 ((void *)nhdr64) + nhdr_size64 +
575 QEMU_ALIGN_UP(nhdr_namesz, phdr_align);
576
577 pvh_start_addr = *elf_note_data_addr;
578 } else {
579 struct elf32_note *nhdr32 = (struct elf32_note *)arg1;
580 uint32_t nhdr_size32 = sizeof(struct elf32_note);
581 uint32_t phdr_align = *(uint32_t *)arg2;
582 uint32_t nhdr_namesz = nhdr32->n_namesz;
583
584 elf_note_data_addr =
585 ((void *)nhdr32) + nhdr_size32 +
586 QEMU_ALIGN_UP(nhdr_namesz, phdr_align);
587
588 pvh_start_addr = *(uint32_t *)elf_note_data_addr;
589 }
590
591 return pvh_start_addr;
592 }
593
load_elfboot(const char * kernel_filename,int kernel_file_size,uint8_t * header,size_t pvh_xen_start_addr,FWCfgState * fw_cfg)594 static bool load_elfboot(const char *kernel_filename,
595 int kernel_file_size,
596 uint8_t *header,
597 size_t pvh_xen_start_addr,
598 FWCfgState *fw_cfg)
599 {
600 uint32_t flags = 0;
601 uint32_t mh_load_addr = 0;
602 uint32_t elf_kernel_size = 0;
603 uint64_t elf_entry;
604 uint64_t elf_low, elf_high;
605 int kernel_size;
606
607 if (ldl_le_p(header) != 0x464c457f) {
608 return false; /* no elfboot */
609 }
610
611 bool elf_is64 = header[EI_CLASS] == ELFCLASS64;
612 flags = elf_is64 ?
613 ((Elf64_Ehdr *)header)->e_flags : ((Elf32_Ehdr *)header)->e_flags;
614
615 if (flags & 0x00010004) { /* LOAD_ELF_HEADER_HAS_ADDR */
616 error_report("elfboot unsupported flags = %x", flags);
617 exit(1);
618 }
619
620 uint64_t elf_note_type = XEN_ELFNOTE_PHYS32_ENTRY;
621 kernel_size = load_elf(kernel_filename, read_pvh_start_addr,
622 NULL, &elf_note_type, &elf_entry,
623 &elf_low, &elf_high, NULL,
624 ELFDATA2LSB, I386_ELF_MACHINE, 0, 0);
625
626 if (kernel_size < 0) {
627 error_report("Error while loading elf kernel");
628 exit(1);
629 }
630 mh_load_addr = elf_low;
631 elf_kernel_size = elf_high - elf_low;
632
633 if (pvh_start_addr == 0) {
634 error_report("Error loading uncompressed kernel without PVH ELF Note");
635 exit(1);
636 }
637 fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ENTRY, pvh_start_addr);
638 fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ADDR, mh_load_addr);
639 fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_SIZE, elf_kernel_size);
640
641 return true;
642 }
643
x86_load_linux(X86MachineState * x86ms,FWCfgState * fw_cfg,int acpi_data_size,bool pvh_enabled)644 void x86_load_linux(X86MachineState *x86ms,
645 FWCfgState *fw_cfg,
646 int acpi_data_size,
647 bool pvh_enabled)
648 {
649 bool linuxboot_dma_enabled = X86_MACHINE_GET_CLASS(x86ms)->fwcfg_dma_enabled;
650 uint16_t protocol;
651 int setup_size, kernel_size, cmdline_size;
652 int dtb_size, setup_data_offset;
653 uint32_t initrd_max;
654 uint8_t header[8192], *setup, *kernel;
655 hwaddr real_addr, prot_addr, cmdline_addr, initrd_addr = 0;
656 FILE *f;
657 char *vmode;
658 MachineState *machine = MACHINE(x86ms);
659 struct setup_data *setup_data;
660 const char *kernel_filename = machine->kernel_filename;
661 const char *initrd_filename = machine->initrd_filename;
662 const char *dtb_filename = machine->dtb;
663 const char *kernel_cmdline = machine->kernel_cmdline;
664 SevKernelLoaderContext sev_load_ctx = {};
665
666 /* Align to 16 bytes as a paranoia measure */
667 cmdline_size = (strlen(kernel_cmdline) + 16) & ~15;
668
669 /* load the kernel header */
670 f = fopen(kernel_filename, "rb");
671 if (!f) {
672 fprintf(stderr, "qemu: could not open kernel file '%s': %s\n",
673 kernel_filename, strerror(errno));
674 exit(1);
675 }
676
677 kernel_size = get_file_size(f);
678 if (!kernel_size ||
679 fread(header, 1, MIN(ARRAY_SIZE(header), kernel_size), f) !=
680 MIN(ARRAY_SIZE(header), kernel_size)) {
681 fprintf(stderr, "qemu: could not load kernel '%s': %s\n",
682 kernel_filename, strerror(errno));
683 exit(1);
684 }
685
686 /*
687 * kernel protocol version.
688 * Please see https://www.kernel.org/doc/Documentation/x86/boot.txt
689 */
690 if (ldl_le_p(header + 0x202) == 0x53726448) /* Magic signature "HdrS" */ {
691 protocol = lduw_le_p(header + 0x206);
692 } else {
693 /*
694 * This could be a multiboot kernel. If it is, let's stop treating it
695 * like a Linux kernel.
696 * Note: some multiboot images could be in the ELF format (the same of
697 * PVH), so we try multiboot first since we check the multiboot magic
698 * header before to load it.
699 */
700 if (load_multiboot(x86ms, fw_cfg, f, kernel_filename, initrd_filename,
701 kernel_cmdline, kernel_size, header)) {
702 return;
703 }
704 /*
705 * Check if the file is an uncompressed kernel file (ELF) and load it,
706 * saving the PVH entry point used by the x86/HVM direct boot ABI.
707 * If load_elfboot() is successful, populate the fw_cfg info.
708 */
709 if (pvh_enabled &&
710 load_elfboot(kernel_filename, kernel_size,
711 header, pvh_start_addr, fw_cfg)) {
712 fclose(f);
713
714 fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE,
715 strlen(kernel_cmdline) + 1);
716 fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA, kernel_cmdline);
717
718 setup = g_memdup2(header, sizeof(header));
719
720 fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_SIZE, sizeof(header));
721 fw_cfg_add_bytes(fw_cfg, FW_CFG_SETUP_DATA,
722 setup, sizeof(header));
723
724 /* load initrd */
725 if (initrd_filename) {
726 GMappedFile *mapped_file;
727 gsize initrd_size;
728 gchar *initrd_data;
729 GError *gerr = NULL;
730
731 mapped_file = g_mapped_file_new(initrd_filename, false, &gerr);
732 if (!mapped_file) {
733 fprintf(stderr, "qemu: error reading initrd %s: %s\n",
734 initrd_filename, gerr->message);
735 exit(1);
736 }
737 x86ms->initrd_mapped_file = mapped_file;
738
739 initrd_data = g_mapped_file_get_contents(mapped_file);
740 initrd_size = g_mapped_file_get_length(mapped_file);
741 initrd_max = x86ms->below_4g_mem_size - acpi_data_size - 1;
742 if (initrd_size >= initrd_max) {
743 fprintf(stderr, "qemu: initrd is too large, cannot support."
744 "(max: %"PRIu32", need %"PRId64")\n",
745 initrd_max, (uint64_t)initrd_size);
746 exit(1);
747 }
748
749 initrd_addr = (initrd_max - initrd_size) & ~4095;
750
751 fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_ADDR, initrd_addr);
752 fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_SIZE, initrd_size);
753 fw_cfg_add_bytes(fw_cfg, FW_CFG_INITRD_DATA, initrd_data,
754 initrd_size);
755 }
756
757 option_rom[nb_option_roms].bootindex = 0;
758 option_rom[nb_option_roms].name = "pvh.bin";
759 nb_option_roms++;
760
761 return;
762 }
763 protocol = 0;
764 }
765
766 if (protocol < 0x200 || !(header[0x211] & 0x01)) {
767 /* Low kernel */
768 real_addr = 0x90000;
769 cmdline_addr = 0x9a000 - cmdline_size;
770 prot_addr = 0x10000;
771 } else if (protocol < 0x202) {
772 /* High but ancient kernel */
773 real_addr = 0x90000;
774 cmdline_addr = 0x9a000 - cmdline_size;
775 prot_addr = 0x100000;
776 } else {
777 /* High and recent kernel */
778 real_addr = 0x10000;
779 cmdline_addr = 0x20000;
780 prot_addr = 0x100000;
781 }
782
783 /* highest address for loading the initrd */
784 if (protocol >= 0x20c &&
785 lduw_le_p(header + 0x236) & XLF_CAN_BE_LOADED_ABOVE_4G) {
786 /*
787 * Linux has supported initrd up to 4 GB for a very long time (2007,
788 * long before XLF_CAN_BE_LOADED_ABOVE_4G which was added in 2013),
789 * though it only sets initrd_max to 2 GB to "work around bootloader
790 * bugs". Luckily, QEMU firmware(which does something like bootloader)
791 * has supported this.
792 *
793 * It's believed that if XLF_CAN_BE_LOADED_ABOVE_4G is set, initrd can
794 * be loaded into any address.
795 *
796 * In addition, initrd_max is uint32_t simply because QEMU doesn't
797 * support the 64-bit boot protocol (specifically the ext_ramdisk_image
798 * field).
799 *
800 * Therefore here just limit initrd_max to UINT32_MAX simply as well.
801 */
802 initrd_max = UINT32_MAX;
803 } else if (protocol >= 0x203) {
804 initrd_max = ldl_le_p(header + 0x22c);
805 } else {
806 initrd_max = 0x37ffffff;
807 }
808
809 if (initrd_max >= x86ms->below_4g_mem_size - acpi_data_size) {
810 initrd_max = x86ms->below_4g_mem_size - acpi_data_size - 1;
811 }
812
813 fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_ADDR, cmdline_addr);
814 fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE, strlen(kernel_cmdline) + 1);
815 fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA, kernel_cmdline);
816 sev_load_ctx.cmdline_data = (char *)kernel_cmdline;
817 sev_load_ctx.cmdline_size = strlen(kernel_cmdline) + 1;
818
819 if (protocol >= 0x202) {
820 stl_le_p(header + 0x228, cmdline_addr);
821 } else {
822 stw_le_p(header + 0x20, 0xA33F);
823 stw_le_p(header + 0x22, cmdline_addr - real_addr);
824 }
825
826 /* handle vga= parameter */
827 vmode = strstr(kernel_cmdline, "vga=");
828 if (vmode) {
829 unsigned int video_mode;
830 const char *end;
831 int ret;
832 /* skip "vga=" */
833 vmode += 4;
834 if (!strncmp(vmode, "normal", 6)) {
835 video_mode = 0xffff;
836 } else if (!strncmp(vmode, "ext", 3)) {
837 video_mode = 0xfffe;
838 } else if (!strncmp(vmode, "ask", 3)) {
839 video_mode = 0xfffd;
840 } else {
841 ret = qemu_strtoui(vmode, &end, 0, &video_mode);
842 if (ret != 0 || (*end && *end != ' ')) {
843 fprintf(stderr, "qemu: invalid 'vga=' kernel parameter.\n");
844 exit(1);
845 }
846 }
847 stw_le_p(header + 0x1fa, video_mode);
848 }
849
850 /* loader type */
851 /*
852 * High nybble = B reserved for QEMU; low nybble is revision number.
853 * If this code is substantially changed, you may want to consider
854 * incrementing the revision.
855 */
856 if (protocol >= 0x200) {
857 header[0x210] = 0xB0;
858 }
859 /* heap */
860 if (protocol >= 0x201) {
861 header[0x211] |= 0x80; /* CAN_USE_HEAP */
862 stw_le_p(header + 0x224, cmdline_addr - real_addr - 0x200);
863 }
864
865 /* load initrd */
866 if (initrd_filename) {
867 GMappedFile *mapped_file;
868 gsize initrd_size;
869 gchar *initrd_data;
870 GError *gerr = NULL;
871
872 if (protocol < 0x200) {
873 fprintf(stderr, "qemu: linux kernel too old to load a ram disk\n");
874 exit(1);
875 }
876
877 mapped_file = g_mapped_file_new(initrd_filename, false, &gerr);
878 if (!mapped_file) {
879 fprintf(stderr, "qemu: error reading initrd %s: %s\n",
880 initrd_filename, gerr->message);
881 exit(1);
882 }
883 x86ms->initrd_mapped_file = mapped_file;
884
885 initrd_data = g_mapped_file_get_contents(mapped_file);
886 initrd_size = g_mapped_file_get_length(mapped_file);
887 if (initrd_size >= initrd_max) {
888 fprintf(stderr, "qemu: initrd is too large, cannot support."
889 "(max: %"PRIu32", need %"PRId64")\n",
890 initrd_max, (uint64_t)initrd_size);
891 exit(1);
892 }
893
894 initrd_addr = (initrd_max - initrd_size) & ~4095;
895
896 fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_ADDR, initrd_addr);
897 fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_SIZE, initrd_size);
898 fw_cfg_add_bytes(fw_cfg, FW_CFG_INITRD_DATA, initrd_data, initrd_size);
899 sev_load_ctx.initrd_data = initrd_data;
900 sev_load_ctx.initrd_size = initrd_size;
901
902 stl_le_p(header + 0x218, initrd_addr);
903 stl_le_p(header + 0x21c, initrd_size);
904 }
905
906 /* load kernel and setup */
907 setup_size = header[0x1f1];
908 if (setup_size == 0) {
909 setup_size = 4;
910 }
911 setup_size = (setup_size + 1) * 512;
912 if (setup_size > kernel_size) {
913 fprintf(stderr, "qemu: invalid kernel header\n");
914 exit(1);
915 }
916
917 setup = g_malloc(setup_size);
918 kernel = g_malloc(kernel_size);
919 fseek(f, 0, SEEK_SET);
920 if (fread(setup, 1, setup_size, f) != setup_size) {
921 fprintf(stderr, "fread() failed\n");
922 exit(1);
923 }
924 fseek(f, 0, SEEK_SET);
925 if (fread(kernel, 1, kernel_size, f) != kernel_size) {
926 fprintf(stderr, "fread() failed\n");
927 exit(1);
928 }
929 fclose(f);
930
931 /* append dtb to kernel */
932 if (dtb_filename) {
933 if (protocol < 0x209) {
934 fprintf(stderr, "qemu: Linux kernel too old to load a dtb\n");
935 exit(1);
936 }
937
938 dtb_size = get_image_size(dtb_filename);
939 if (dtb_size <= 0) {
940 fprintf(stderr, "qemu: error reading dtb %s: %s\n",
941 dtb_filename, strerror(errno));
942 exit(1);
943 }
944
945 setup_data_offset = QEMU_ALIGN_UP(kernel_size, 16);
946 kernel_size = setup_data_offset + sizeof(struct setup_data) + dtb_size;
947 kernel = g_realloc(kernel, kernel_size);
948
949 stq_le_p(header + 0x250, prot_addr + setup_data_offset);
950
951 setup_data = (struct setup_data *)(kernel + setup_data_offset);
952 setup_data->next = 0;
953 setup_data->type = cpu_to_le32(SETUP_DTB);
954 setup_data->len = cpu_to_le32(dtb_size);
955
956 load_image_size(dtb_filename, setup_data->data, dtb_size);
957 }
958
959 /*
960 * If we're starting an encrypted VM, it will be OVMF based, which uses the
961 * efi stub for booting and doesn't require any values to be placed in the
962 * kernel header. We therefore don't update the header so the hash of the
963 * kernel on the other side of the fw_cfg interface matches the hash of the
964 * file the user passed in.
965 */
966 if (!sev_enabled() && protocol > 0) {
967 memcpy(setup, header, MIN(sizeof(header), setup_size));
968 }
969
970 fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ADDR, prot_addr);
971 fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_SIZE, kernel_size - setup_size);
972 fw_cfg_add_bytes(fw_cfg, FW_CFG_KERNEL_DATA,
973 kernel + setup_size, kernel_size - setup_size);
974 sev_load_ctx.kernel_data = (char *)kernel + setup_size;
975 sev_load_ctx.kernel_size = kernel_size - setup_size;
976
977 fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_ADDR, real_addr);
978 fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_SIZE, setup_size);
979 fw_cfg_add_bytes(fw_cfg, FW_CFG_SETUP_DATA, setup, setup_size);
980 sev_load_ctx.setup_data = (char *)setup;
981 sev_load_ctx.setup_size = setup_size;
982
983 /* kernel without setup header patches */
984 fw_cfg_add_file(fw_cfg, "etc/boot/kernel", kernel, kernel_size);
985
986 if (machine->shim_filename) {
987 GMappedFile *mapped_file;
988 GError *gerr = NULL;
989
990 mapped_file = g_mapped_file_new(machine->shim_filename, false, &gerr);
991 if (!mapped_file) {
992 fprintf(stderr, "qemu: error reading shim %s: %s\n",
993 machine->shim_filename, gerr->message);
994 exit(1);
995 }
996
997 fw_cfg_add_file(fw_cfg, "etc/boot/shim",
998 g_mapped_file_get_contents(mapped_file),
999 g_mapped_file_get_length(mapped_file));
1000 }
1001
1002 if (sev_enabled()) {
1003 sev_add_kernel_loader_hashes(&sev_load_ctx, &error_fatal);
1004 }
1005
1006 option_rom[nb_option_roms].bootindex = 0;
1007 option_rom[nb_option_roms].name = "linuxboot.bin";
1008 if (linuxboot_dma_enabled && fw_cfg_dma_enabled(fw_cfg)) {
1009 option_rom[nb_option_roms].name = "linuxboot_dma.bin";
1010 }
1011 nb_option_roms++;
1012 }
1013
x86_isa_bios_init(MemoryRegion * isa_bios,MemoryRegion * isa_memory,MemoryRegion * bios,bool read_only)1014 void x86_isa_bios_init(MemoryRegion *isa_bios, MemoryRegion *isa_memory,
1015 MemoryRegion *bios, bool read_only)
1016 {
1017 uint64_t bios_size = memory_region_size(bios);
1018 uint64_t isa_bios_size = MIN(bios_size, 128 * KiB);
1019
1020 memory_region_init_alias(isa_bios, NULL, "isa-bios", bios,
1021 bios_size - isa_bios_size, isa_bios_size);
1022 memory_region_add_subregion_overlap(isa_memory, 1 * MiB - isa_bios_size,
1023 isa_bios, 1);
1024 memory_region_set_readonly(isa_bios, read_only);
1025 }
1026
x86_bios_rom_init(X86MachineState * x86ms,const char * default_firmware,MemoryRegion * rom_memory,bool isapc_ram_fw)1027 void x86_bios_rom_init(X86MachineState *x86ms, const char *default_firmware,
1028 MemoryRegion *rom_memory, bool isapc_ram_fw)
1029 {
1030 const char *bios_name;
1031 char *filename;
1032 int bios_size;
1033 ssize_t ret;
1034
1035 /* BIOS load */
1036 bios_name = MACHINE(x86ms)->firmware ?: default_firmware;
1037 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1038 if (filename) {
1039 bios_size = get_image_size(filename);
1040 } else {
1041 bios_size = -1;
1042 }
1043 if (bios_size <= 0 ||
1044 (bios_size % 65536) != 0) {
1045 goto bios_error;
1046 }
1047 if (machine_require_guest_memfd(MACHINE(x86ms))) {
1048 memory_region_init_ram_guest_memfd(&x86ms->bios, NULL, "pc.bios",
1049 bios_size, &error_fatal);
1050 if (is_tdx_vm()) {
1051 tdx_set_tdvf_region(&x86ms->bios);
1052 }
1053 } else {
1054 memory_region_init_ram(&x86ms->bios, NULL, "pc.bios",
1055 bios_size, &error_fatal);
1056 }
1057 if (sev_enabled() || is_tdx_vm()) {
1058 /*
1059 * The concept of a "reset" simply doesn't exist for
1060 * confidential computing guests, we have to destroy and
1061 * re-launch them instead. So there is no need to register
1062 * the firmware as rom to properly re-initialize on reset.
1063 * Just go for a straight file load instead.
1064 */
1065 void *ptr = memory_region_get_ram_ptr(&x86ms->bios);
1066 load_image_size(filename, ptr, bios_size);
1067 x86_firmware_configure(0x100000000ULL - bios_size, ptr, bios_size);
1068 } else {
1069 memory_region_set_readonly(&x86ms->bios, !isapc_ram_fw);
1070 ret = rom_add_file_fixed(bios_name, (uint32_t)(-bios_size), -1);
1071 if (ret != 0) {
1072 goto bios_error;
1073 }
1074 }
1075 g_free(filename);
1076
1077 if (!machine_require_guest_memfd(MACHINE(x86ms))) {
1078 /* map the last 128KB of the BIOS in ISA space */
1079 x86_isa_bios_init(&x86ms->isa_bios, rom_memory, &x86ms->bios,
1080 !isapc_ram_fw);
1081 }
1082
1083 /* map all the bios at the top of memory */
1084 memory_region_add_subregion(rom_memory,
1085 (uint32_t)(-bios_size),
1086 &x86ms->bios);
1087 return;
1088
1089 bios_error:
1090 fprintf(stderr, "qemu: could not load PC BIOS '%s'\n", bios_name);
1091 exit(1);
1092 }
1093